• quasars: general;
  • galaxies: structure;
  • gravitational lensing.

We discuss whether one should expect that multiply imaged QSOs can be understood with ‘simple’ lens models that contain only a few parameters. Whereas for many lens systems such simple mass models yield a remarkably good description of the observed properties, there are some systems which are notoriously difficult to understand quantitatively. We argue that at least in one case (B 1422+231) these difficulties are not (solely) due to a ‘wrong’ parametrization of the lens model, but that the discrepancy between observed and model-predicted flux ratios is due to substructure in the lens. As in microlensing for optical fluxes, such substructure can distort also the radio flux ratios predicted by ‘simple’ mass models, in particular for highly magnified images, without appreciably changing image positions. Substructure also does not change the time delay significantly, and therefore has little effect on the determination of the Hubble constant using time delays. We quantify these statements with several simple scenarios for substructure, and propose a strategy to model lens systems in which substructure is suspected.